A more perfect unification exploring a nano syntactic solution to vietnamese

DOI: 10.1075/slcs.211.04pha Link: https://benjamins.com/catalog/slcs.211.04pha A more perfect unification: exploring a Nano-syntactic solution to Vietnamese đã Trang Phan University of

1

Please cite it as: Trang Phan & Nigel Duffield 2019 A more perfect unification:

exploring a Nano-syntactic solution to Vietnamese đã In: N Duffield, T Phan &

T Trinh (eds.), Interdisciplinary Perspectives on Vietnamese Linguistics Studies in

Language Companion Series 211, pp 69-80 John Benjamins

DOI: 10.1075/slcs.211.04pha

Link: https://benjamins.com/catalog/slcs.211.04pha

A more perfect unification: exploring a Nano-syntactic

solution to Vietnamese đã

Trang Phan

University of Languages & International Studies, Vietnam National University –

Hanoi, Vietnam

Nigel Duffield

Konan University, Japan

Abstract

In this paper, we provide a new analysis of the Negative Constraint in Vietnamese,

whereby the anterior morpheme đã loses its perfect reading in negative contexts

The Nanosyntax approach adopted here is claimed to derive this constraint without the stipulations inherent in existing formal accounts (e.g., Trinh 2005, Phan & Duffield (2016, 2017)

Keywords

Vietnamese syntax, Aspect-Negation interactions, Negative Constraint

1 Introduction

The empirical concern of this paper1 is the contrastive behaviour of the

Vietnamese TAM marker đã across affirmative contexts vs negative contexts

Specifically, our concern is with the fact that in affirmative sentences the presence

of đã gives rise to an ambiguity between a past and a perfect reading, whereas in

negative contexts only the preterite reading is available For obvious reasons, we refer to this as the Negative Constraint

Let us first consider some data, beginning with affirmative contexts:

1 The present article is an attempt to improve upon our previous analyses of the Negative Constraint: see, for example, Phan & Duffield (2016, 2017), Phan & Duffield (2018)

2

(1) a Anh-ấy đến

3S.M come

‘He comes/came.’ [No specified time]

b Anh-ấy đã đến

3S.M DA come

EITHER: ‘He came.’ [Past time interpretation]

OR: ‘He has come.’ [Perfect interpretation]

In (1a) – the sentence without đã – the man’s coming may be freely

interpreted as taking place in the present or in the past In (1b), on the other hand,

the presence of đã situates the event in the past However, in addition to this past

time (preterite) reading, (1b) may also be interpreted with a perfect reading; that is

to say, the man’s arrival is asserted to have occurred prior to the utterance time, and still to be of current relevance

Now consider the interaction between đã and clausal negation There are two

negative markers in Vietnamese that we are concerned with in this paper: the simple

negative morpheme không (NEG) and perfect negative chưa (NEG PRF), usually translated as ‘not yet’: these are exemplified in (2) and (3), respectively

(2) a Anh-ấy không đến

3S.M NEG come

‘He doesn’t come/didn’t come.’

3S.M DA NEG come

‘He didn’t come.’ [exclusive past time interpretation]

NOT ‘He hasn’t come.’

(3) a Anh-ấy chưa đến

3S.M NEG PRF come

‘He hasn’t come yet.’ [exclusive perfect interpretation]

3S.M DA NEG PRF come

‘He hadn’t come yet.’ [past perfect interpretation]

Simple negative sentences, such as the example in (2a) – which contain không but without đã, – are compatible with either a present or a past time interpretation Addition of đã to a negative clause, as in the example (2b), yields a past time

interpretation only: the perfect reading is excluded here In order to obtain a

negative perfect reading the default negative không in (2) must be replaced by the synthetic negative marker chưa (‘not.yet’), illustrated in (3a) and (3b) Where this form appears on its own, as in (3a), chưa has an exclusively perfect reading; that is

to say, it cannot be used to indicate a definite time in the past The addition of đã in

(3b) immediately shifts the interpretation from a present perfect to a past perfect

one This clearly suggests that the sole interpretive contribution of đã in negative

sentences is to add a past time reading

While these observations concerning đã have been previously discussed in the literature – see e.g., Panfilov (2002), Trinh (2005); Duffield (2013, 2014, 2017),

Phan (2013), Bui (this volume) – no completely satisfactory explanation has yet emerged of the Negative Constraint The aim of this squib is to sketch out an

original syntactic approach to đã using the Spell-out principles of Nanosyntax: we

3

shall claim that the advantages this approach offers over earlier head-movement-driven accounts makes it the most promising to date

Before continuing, it is worth noting that this kind of interaction between aspect and negation is not unique to Vietnamese: it has previously been observed that certain kinds of aspectual reading may appear or disappear in negative

contexts; cf Matthews (1990), Li (1999), Miestamo et al (2011)

2 Previous Treatments

Hitherto, there have been two main syntactic approaches to the Negation

Constraint: the ‘đã-as-homophone’ analysis, as proposed by Trinh (2005), and the

‘multifunctional-đã’ approach advanced by Duffield (2013, 2014) and Phan (2013)

We briefly review these in turn

Trinh (2005)’s account tackles the problem by assuming that there are two

homophonous lexical items: ‘perfect ĐÃ 1 ’ and ‘past ĐÃ 2 ’, each having different

points of initial merger Specifically, the perfect ĐÃ 1 is initially merged lower in Aspo, then raises to T0 yielding the ambiguous interpretation of đã in (1b), as shown

in (4a) By contrast, past ĐÃ 2 is taken to be directly base-generated in To, as illustrated in (4b): this yields the exclusive past interpretation in (2b) and (3b) above

(4) a Affirmative clauses (OAspàT)

b Negative clauses (direct insertion under T)

4

There are two significant difficulties with Trinh’s account In the first place, it fails to capture the close semantic relationship between perfect and past readings: it

is presumably not accidental that these meanings are conflated in many languages, such as in many modern varieties of spoken Romance and Continental West Germanic, where preterite forms have largely been lost (either restricted to literary registers, or lost entirely, in some varieties)

More significantly perhaps, Trinh’s analysis offers no explanation – other than possibly though appeal to haplology – as to why these two homophones may not co-occur, either in affirmative or negative contexts (as in (5) and (6), respectively):

(5) *Anh-ấy đã đã đến

3S.M DA DA come

‘He came’/‘He has come.’

(6) a *Anh-ấy đã không đã đến

3S.M DA NEG DA come

‘He didn’t come.’

3S.M DA NEG PRF DA come

‘He hadn’t come yet.’

The analysis proposed in Duffield (2013, 2014), also Phan (2013), is in many respects a variant of Trinh’s account It avoids the problem of accidental homophony by invoking the notion of multifunctionality in the sense of Travis, Bobaljik & Lefebvre (1998), Duffield (2014), according to which grammatical meaning inheres in syntactic heads themselves, rather than in the underspecified lexical exponents of these heads On the original Duffield/Phan account, there is

only one lexical đã: its interpretation in a given context is the sum of its core

meaning – namely, ‘anterior’ – and whatever additional meanings it derives from

the grammatical positions into which it is merged Thus, đã is ambiguous if it is

first merged under Asp° and later raised to T°, but is unambiguous – signalling the past-only reading – whenever it is directly inserted under T°

This multifunctional approach nicely captures the intuition that different

5

interpretations of đã result from different syntactic environments, and directly explains the absence of doubled đã in affirmative contexts, as in (5) above

However, it leaves unresolved the question of why negative sentences such as those

in (6) are unacceptable even though both positions – above and below Neg° – should be available

Both previous analyses trace the Negation Constraint to the idea that the presence of negation triggers a violation of head-minimality: as with tense-lowering

in English (Pollock 1989, Chomsky 1989), clausal negation is assumed to block head-movement Yet though the analogy is obvious, it is much less clear why negation should block Asp-to-T raising here; after all, finite auxiliary raising over negation would seem to be the rule rather than the exception in more familiar languages This putative blocking effect is particularly puzzling since there are no morphosyntactic considerations – ‘Mirror Principle effects’ – that would require strict adherence to the Head Movement Constraint (Travis 1984): in the case at hand, long head-movement should be permissible; see Harizanov & Gribanova (2018), for a revised approach to the HMC Given this, we are led to consider an alternative approach to the negation puzzle: rather than invoking head-movement,

we propose to explain the Negation Constraint in terms of competition among

lexical – or rather lexico-syntactic – items {đã, chưa, không} when it comes to spelling out the syntactic structure T > Neg > Asp The present account relies on the

lexicalization algorithms of Nanosyntax, which are set out in the next section

3 A Nanosyntactic approach to the Negation Constraint

Following Starke (2009, 2011), Caha (2009), and Lander (2016), we assume that words, including functional categories, are lexically represented as L-TREES, which may – if they are complex — correspond to a continuous stretch of syntactic phrase-structure, that is to say, a word corresponds to more than a single syntactic head On this construal, where we have a syntactic tree (S-TREE) that needs to be spelled out, it is necessary to match all available L-trees to the S-tree.2 There will be

a competition between those L-trees: which competitor wins out in this mapping contest in a given context is determined by three governing principles:

• SUPERSET PRINCIPLE, which requires that an L-tree should be the same size

or larger than the relevant S-tree for a successful match, see Caha (2009, 2014);

• ELSEWHERE PRINCIPLE, which requires that – just in case more than a single L-tree is available to lexicalise an S-tree – the L-tree with the fewest unused features should be chosen;

• PRINCIPLE OF CYCLIC OVERRIDE: assuming that derivations are built

bottom-up, then later, higher-level outs cancel out previous, lower-level spell-outs: see Lander (2016), for discussion

2 Within the assumptions of Nanosyntax, the computation starts from features The syntax does not project from lexical items (as is more commonly assumed), but rather the other way around The core idea in Nanosyntax is that the lexicon is strictly post-syntactic: there is no pre-syntactic lexicon, as in Minimalism, nor or there 'lists' that feed into syntax, as proposed in Distributed Morphology The justification behind this kind of architecture ultimately has to do with the idea of sub-morphemic heads, and the need for phrasal spellout See Baunaz, De Clercq, Haegeman & Lander (2018), for explication and detailed discussion We are grateful to a reviewer for raising this issue

6

Here, we adopt a compositional approach to đã – for fuller justification see Phan & Duffield (2018) – in which đã is taken to comprise two semantic features: a

temporal PAST feature, and an aspectual PERFECT feature Syntactically, these two features head their own strictly-ordered projections: PastP > PerfectP In terms of Nanosyntax, ĐÃ is a lexico-syntactic object, instantiated as the layered L-tree in (7):

(7)

Þ ĐÃ

The Vietnamese lexicon also contains two abstract lexical items, KHÔNG and

CHƯA: these are associated with the L-trees in (8) and (9), respectively

(8) L-tree for KHÔNG

Þ KHÔNG

(9) L-tree for CHƯA

Þ CHƯA

Crucial to the present analysis is the top layer InterrogativeP, above NegP in

(8) and (9). 3 This additional structure reflects the fact that không and chưa can

serve as either negative or interrogative markers, depending on their position with

respect to VP; cf Duffield (2013), Trinh (2005), Law (2014), for further discussion Intuitvely then, không = (yes or) no; whereas chưa = (yes or) not yet: cf Nguyen D

H (1997) These extended trees for không and chưa in (8) and (9) thus minimally

contrast with what is proposed in Duffield (2017: tree 20), in which the L-tree for

3 We are grateful to Lena Baunaz and to Karen De Clercq for discussion of this point

7

chưa contained only two layers (NegP>PerfP) On that earlier proposal đã would

have no fewer unused features than chưa when it comes to spelling out PerfectP,

and so could not be preferred by the Elsewhere Principle

By the Superset Principle,4 the L-tree in (7) can match all of the S-trees in

(10), yielding the multifunctional ambiguity effect of đã: 5

(10) a S1: Past-Perfect

b S2: Past

c S3: Perfect

Syntax incorporates one feature at a time, and at each step, a suitable match

from the lexicon must be found First let us consider affirmative perfect contexts, as

in the ambiguous examples (1) and (*5) above In affirmative perfect sentences, we have only PerfectP There are two L-tree candidates that can lexicalise PerfectP —

either đã or chưa — since by the Superset Principle, PerfectP is contained both in the L-tree for đã in (7) and also in the L-tree for chưa in (9) Given the Elsewhere

4 In this squib, we adopt the revised version of the Superset Principle, following Caha (2014), which is the original version (Stark 2009, Lander 2016) without the ‘Anchor condition’ The crucial difference between the two versions is that the revised version allows the L-tree to spell out all three S-trees (a) (b) and (c) in (10); whereas in the classical version only (a) and (c) are allowed, (b) is not allowed, since the lowest layer PerfectP has to be matched by the Anchor condition; see Lander (2016), for details, see also De Clercq & Vanden Wyngaerd (2016), Vanden Wyngaerd (2016), for further discussion of the revised Superset Principle can account for other grammatical phenomena cross-linguistically We are grateful to Amélie Rocquet, Pavel Caha, Eric Lander and Karen De Clercq for discussing this point

5 One anonymous reviewer raises the question of whether the so-called mapping between the lexical syntax and the genuine syntactic structure is nothing more than a different way to maintain the “homophone” approach, only to shift part of the burden to the syntactic structure As will be shown below, Nanosyntax is certainly not a different way of maintaining the homophone approach This is exactly the point of the Superset Principle as a way to account for syncretism See Baunaz, De Clercq, Haegeman & Lander (2018), for detailed discussion However, we suppose in one way we are 'shifting the burden' to syntax - since we assume that different readings require different underlying structures, with a single lexical entry potentially being able to match different sizes

of that syntactic structure by the Superset Principle

8

Principle, the L-tree for đã in (7) is the winning match since it has fewer unused features Accordingly, PerfectP is lexicalised as đã

(11) Affirmative Perfect Derivation (one step):

Þ đã [chưa is the losing competitor]

Þ đã [chưa is losing competitor]

In the case of affirmative past contexts, two derivational steps are involved

First, we start once more with PerfectP: once again, the best match in the lexicon is

the L-tree of đã in (7), so PerfectP is spelled out by đã, as before At the second step

in the derivation, however, when we build PastP on top of PerfectP, there is a match

for the whole trunk PastP>PerfectP in the lexicon, spelled out by đã, which overrides the first spellout The unattested order *đã đã is ungrammatical (example

5) due to the Principle of Cyclic Override

(12) Affirmative Past Derivation (two steps):

Step 1:

Þ đã [chưa is the losing competitor]

Step 2:

Þ đã

Þ đã [deleted by Cyclic Override]

Now consider negative contexts, as in the examples in (2), (3) and (6) In past

– that is non-perfect – negative không sentences, such as example (2b), two

derivational steps are once again involved The first step begins from NegP: here,

the Superset Principle allows for two possible spell-outs – không or chưa – since both the L-tree for không in (8) as well as the L-tree for chưa in (9) are supersets of the S-tree NegP However, the L-tree for không in (8) contains the fewer unused features, so NegP spells out as không, given the Elsewhere Principle At Step 2,

PastP is built on top of NegP At this point there is no match for the whole trunk

PastP>NegP in the lexicon, so NegP is spelled-out by không, while PastP is spelled out by đã, the two independently of one other We end up with the correct word order – that is to say, đã precedes không – and with the desired interpretation, in that đã is interpreted as past only

9

(13) Negative Past Derivation (two steps):

Step 1

Þ không [by Superset Principle]

Step 2

Þ đã

Þ không [from Step 1]

In the case of past perfect negative chưa contexts, as in example (3b) above,

the derivation now involves three steps At Step 1, we start from PerfectP; here, the

best match in the lexicon is the L-tree of đã in (7), so PerfectP is spelled out by đã

At step 2, the derivation proceeds, with NegP being inserted on top of PerfectP

This time there is a lexical (L-syntactic) match for the NegP> PerfectP, namely,

chưa, the L-tree for chưa in (9) being a superset of the S-tree NegP > PerfectP)

This higher spell-out chưa cancels out the previous spell-out (đã); by Cyclic Override, the order *chưa – đã is ruled correctly ruled out Finally, at Step 3, PastP

is built on top of NegP>PerfectP At this point in the derivation, there is no match for the whole trunk in the lexicon, hence only one possibility is permitted:

NegP>PerfP is spelled out by chưa, and PastP is spelled out by đã We end up with the right word order – đã preceding chưa – and with the correct interpretation: only the past reading of đã is available here Once more, the grammatically unacceptable order *đã chưa đã, in (5b), is excluded by the Cyclic Override principle

(14) (Past) Perfect Derivation: 6

Step 1 Perfect Derivation (as in 11):

Þ đã [chưa is the losing competitor]

6 Perfect sentences are derived by applying Steps 1 and 2 only; Past perfect sentences involve the additional Step 3

10

Step 2 Negative Perfect derivation (only one competitor)

Þ chưa

Þ đã [deleted by Cyclic Override]

Step 3 Past Negative Perfect derivation (additive — no available L-tree)

Þ đã

Þ chưa

Þ đã [deleted by Cyclic Override]

By means of these lexicalization algorithms, we end up both with the desired

word order, in as much as đã always precedes negation morphemes - correctly blocking the unacceptable combinations of *đã đã and *đã chưa đã — and with the observed interpretation, correctly excluding the perfect reading of đã in negative

contexts The Negative Constraint, which had previously been a stipulation, now emerges as a theorem

4 Conclusion

In conclusion, our Nanosyntactic approach has several advantages over the previous head-movement driven accounts It is lexically non-redundant, in

assuming only one lexical entry for đã Given that multifunctionality is ubiquitous

in the Vietnamese lexicon (see Duffield 1998, 1999, Phan 2013, Duffield 2014), a Nanosyntax solution prevents massive lexical homophony; at the same time, it is syntactically flexible, in allowing for a single L-tree to match more than one S-tree Most relevantly of course, it correctly derives the Negation Constraint

Finally, as noted at the outset, the Negation Constraint is not restricted to

Vietnamese To take one example, Mandarin Chinese le is also ambiguous between

a temporal and an aspectual reading (Lin 2005), and is incompatible with negation

markers bu and meiyou:

(15) a ta qu le faguo

3S go LE France

‘He went to France.’/‘He has been to France.’

b *ta bu qu le faguo

3S NEG go LE France

‘He did not go to France.’ [Examples from Li 1999: 235]